Microframes

What is a microframe?#

A microframe reifies a binary relation as a frame: the relation itself becomes a first-class frame that can be named, evoked by words, and connected to other frames. This lets us model how entities interact with the same machinery FrameNet already uses for situations.

Structure#

Microframes share the structural foundation of traditional frames, with one constraint: they contain exactly two Frame Elements (FEs) — the Domain and the Range of the relation. Because a microframe is a binary relation, its two FEs are simply that relation's two arguments. Each FE is named to reflect the specific relation it represents, following the usual frame-naming principles.

Features#

Lexical Units#

Like traditional frames, microframes can be evoked by Lexical Units (LUs), letting specific words trigger the relational concepts they encode and bridging lexical semantics with structural relations.

Specificity#

Microframes express much more specific relations than standard FrameNet frames. Where traditional frames model broad conceptual situations, microframes capture precise relational nuances between entities.

Multi-Frame Associations#

A single microframe can be associated with multiple frames, functioning as a reusable semantic component across conceptual domains.

Classification#

Every microframe — regardless of how it is used — carries a single semantic category: the one user-facing axis for browsing, searching, and representing microframes. The categories are type instances (typ_microframe_category, idType 42); a microframe is also free to carry more than one at once, since different tags can answer different questions about the same relation (see the ISL^FOL block below).

Eighteen categories are live today (checked directly against the database, not just the docs). Thirteen are this resource's own long-standing vocabulary:

Natural has been retired — no microframe carries it, and no typ_microframe_category row for it exists in the database at all anymore. Its territory (non-agentive natural causation) is now covered by Causal, plus Force below where a genuine force is actually manifested.

Five further categories come from ISL^FOL (Hedblom, Image Schemas and Concept Invention), added alongside the thirteen above rather than replacing any of them — a microframe can carry one of these and one of the thirteen at once, each answering a different question about the same relation:

agentively_causes, for instance, carries Agentive, Causal, and Force all at once.

Temporal (LTL-style sequencing — precedes, directly_precedes, follows, directly_follows) is documented as a planned sixth ISL^FOL category, distinct from Movement because it answers a different question — order of facts, not spatial relation — but no Temporal row exists in the database yet. This is a real gap between the design and the data, not a naming variant; flag it rather than assuming it's already live when checking or minting a temporal-sequencing microframe.

Alongside the category, each microframe records a coarse grounding kind — property-grounded (the relation holds directly between two individuals) or description-grounded (the relation denotes a situation type that classifies its participants). Grounding kind is not a stored axis of its own; it is read off the microframe's grounding edge — its DUL parent. See DUL Grounding for the full treatment.

Where applicable a microframe also keeps its GL quale (Formal, Constitutive, Telic, Agentive) as provenance — a record of which Generative Lexicon quale the relation descends from, not a separate classification. This is the same four-way vocabulary Frame Elements uses for FE qualia types; the two are independent uses of one shared vocabulary, not the same fact recorded twice.

The minting rule, briefly#

A microframe earns its existence by naming a relation that some row genuinely needs written, and that no existing microframe already names — never by naming an arguable relation, and never by a candidate that is really a symmetric relation read from its other argument, a filler restriction on an existing microframe, a composition of two existing microframes, an existing microframe under an unfamiliar name, or a relation words are actually about (which belongs to a lexical frame, not a microframe). An asymmetric relation's converse is the one exception that does mint on its own account (is_proper_part_of beside has_proper_part) — and is linked to its counterpart by rel_inverse_of, not treated as a substitute for minting it. A named composite of several existing microframe rows (RCC-8's TPP as has_proper_part and has_common_boundary together) mints too, linked to its components by rel_expands_to. Two or more microframes can be tagged as mutually exclusive over the same pair via a shared typ_microframe_category value, rather than needing a one-off pairwise relation for each such case. Full detail — the Demand test, the five exclusions, well-formedness (cty_domain/cty_range/cty_head_target) — is governed by docs/conceptual/conceptual_dimension.md §7.6, not restated here.

Use Cases#

What a microframe actually asserts is a relation between two Frame Elements — one aligned to its Domain, one to its Range. This assertion is independent of any base frame's own semantics or nature: the microframe does not specialize, reify, or re-describe the situation a frame denotes. It simply makes explicit a relation that holds between two FEs. The same microframe abstraction is applied in two distinct ways, by two distinct relation types, doing two genuinely different jobs — rel_structure and rel_tqr are not two names for one mechanism, and neither supersedes the other.

Structural relation between FEs (rel_structure)#

rel_structure makes explicit the semantics between two Frame Elements, named by the microframe that mediates them — a single ternary edge linking the Domain FE, the Range FE, and the microframe itself. The Domain FE fills the microframe's Domain and the Range FE fills its Range.

The two FEs it relates do not have to belong to the same frame. The underlying table carries no same-frame constraint: the ordinary case is a frame's own internal structure (who exerts, what is affected, what someone experiences — has_as_cause, agentively_causes, is_affected_by_agent, cause_experience, asserted directly between two of a frame's own FEs, with no schema or lexical intermediary at all), but the same device also reaches an FE of a different frame directly — recovering content a frame's own FE inventory has no room for, the same way an LU's external profile denotes an element of another frame's cast (see Lexical Units, "Entities through a relation"). Internal versus external is derived on read from whether the two FEs share a frame, exactly the way lu_profiling already derives internal-vs-external profiling — it is never a second, separately-declared fact. Either way, there is no frame-to-frame edge and no claim about which lexical frame "hosts" the relation — rel_structure is a statement about two participants, never about a base frame's own identity. Full detail: docs/conceptual/conceptual_dimension.md §9.3.

TQR grounding of LU meaning (rel_tqr)#

rel_tqr answers a different question: which lexical frame is the lexicalization home for this microframe's relational content — not which two FEs are related, but which frame, as a whole, is where the relation surfaces in the lexicon. Because a microframe is itself a frame, the FrameNet-native way to record this is a frame-to-frame relation carrying paired FE-to-FE relations underneath it. That is purely the implementation mechanism, not a claim that the microframe's meaning derives from the base frame's meaning. Concretely, a microframe is attached by a dedicated TQR relation (rel_tqr) with two tiers:

• Frame level: the microframe is linked to a base frame whose FEs it relates (e.g. contains ↔ Conter, has_as_colour ↔ Cor, has_kinship ↔ Parentesco, agentively_causes ↔ Ação_transitiva). This link is the carrier for the FE alignment below — not an assertion that the microframe's relational content is a facet of that frame's situation.
• Frame-Element level: the microframe's Domain and Range are aligned to the base frame's Frame Elements, so the relation's two arguments are anchored in concrete participant structure.

The functional difference, stated plainly: rel_structure asserts a relation holds between two specific FEs, full stop — it never says anything about a base frame as a whole, and the two FEs can sit in one frame or in two. rel_tqr asserts that a specific lexical frame is the place a microframe's relational content is lexically grounded, with the FE alignment underneath it only there to anchor that grounding in concrete participant structure. A microframe can participate in both at once, for different purposes — e.g. agentively_causes both mediates an ordinary rel_structure row inside many individual frames' own Agonist×Event structure (§9.3's affectedness mechanism), and separately carries a rel_tqr link to Ação_transitiva as its own lexicalization home.

A purpose-built relation is used because neither standard relation fits: Subframe decomposes a complex/scenario frame into temporally ordered sub-events, and a microframe is a binary relation rather than a sub-event; Perspective re-construes one scene from two full frames with parallel FE inventories, whereas a microframe just relates two FEs. The rel_tqr binding instead records exactly the Domain/Range-to-FE alignment that those relations cannot carry. This is the TQR FE structure detailed in the TQR document, which spells out how the frame-to-frame and FE-to-FE edges are laid down, and how the TQR_LU layer records relations between individual Lexical Units on top of it.

Microframes are also organized among themselves by subsumption (rel_inheritance), and each is grounded in DUL through the same subsumption edge — see TQR Reference and DUL Grounding.

Ontological Relations#

Microframes expresse ontological relations (object properties).

Generalization Across Events#

Microframes also generalize circumstantial and participant dimensions common across many events — Time, Manner, Means, and similar. In TQR these are the non-qualia relations (handled by the Spatial, Role, and General categories), captured with the same microframe abstraction as the qualia relations above.